Aedat #23
Aedat #23
Conversation
python/faery/events_filter.py
Outdated
There was a problem hiding this comment.
I like the vmap-friendly approach. Good one.
python/faery/spectrogram.py
Outdated
There was a problem hiding this comment.
Very very nice addition. We really should add a list of all these great features in the documentation soon...
| @@ -46,23 +46,21 @@ impl Decoder { | |||
| off_value: Vec<u8>, | |||
| skip_errors: bool, | |||
| ) -> PyResult<Self> { | |||
| Python::with_gil(|python| -> PyResult<Self> { | |||
There was a problem hiding this comment.
Why don't you need the GIL anymore?
There was a problem hiding this comment.
I am not 100% sure. I think that pyo3 does some magic under the hood through Rust lifetimes. They mentioned in their changelog that the recent changes to the pyo3 interface were meant to prepare for non-GIL python versions (https://peps.python.org/pep-0703/).
| @@ -55,6 +55,38 @@ enum Function { | |||
| colors: (u32, u32, u32), | |||
| ts_and_polarities: Vec<(u64, neuromorphic_types::DvsPolarity)>, | |||
| }, | |||
|
|
|||
| // activity = activity * exp(-Δt / 𝜏) + 1 | |||
There was a problem hiding this comment.
Probably not but I figured that they might prove useful to someone else. I used them during development to make sure that my implementation. was correct (it can be hard to read since we are computing intermediary values during initialization, which is good for speed but obfuscates the equations a little).
The filter silences pixels whose activity is significantly larger than the most active of their four direct neighbors.
This makes a big visual difference on noisy recordings. Now the default for wiggle plots.
Multiple changes to ensure that generated aedat files can be loaded by DV. The biggest change is the addition of file data tables (essentially a data index at the end of aedat files), which DV requires. Whilst loading and play back works, I've experienced minor issues when trying to load multiple files in a row (restarting DV is often required), when playing short files, or when enabling the loop playback function of DV. Some of these issues could come from the size of the packets generated by faery (typically bigger than DV's), a bug in faery.aedat, or a bug in DV. Since the latter is not completely open-source, we will need help from Inivation / Synsense to investigate further (if there is interest for better support). Fixes #13.
Added the implementation for linear scale (which was missing before) and fixed a few edge cases (in particular constant event rates and no events).
See commits below.